ai-marketing-videos
Pass
Audited by Gen Agent Trust Hub on Aug 31, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill takes user-provided text for video prompts and inserts them into shell commands for the belt CLI. This creates a surface where a malicious user could attempt to break out of the command structure.
- Ingestion points: The skill processes user-defined strings for video and audio generation prompts throughout SKILL.md.
- Boundary markers: There are no explicit boundary markers or instructions to the agent to treat these inputs as untrusted data.
- Capability inventory: The skill uses the Bash tool (restricted to belt commands) to execute the generated prompts.
- Sanitization: No input sanitization or escaping logic is described to prevent command injection or malformed JSON payloads.
- [EXTERNAL_DOWNLOADS]: The skill instructions require the installation of external tools and dependencies from the platform's infrastructure.
- The skill references the installation of 'belt-sh/cli' and multiple 'inference-sh/skills' via npx.
- It links to installation scripts hosted on raw.githubusercontent.com for the platform's CLI.
- [COMMAND_EXECUTION]: The skill relies on executing shell commands using the belt CLI tool to interact with AI models.
- The allowed-tools configuration in the frontmatter restricts the agent to only running commands starting with 'belt', which mitigates the risk of arbitrary command execution.
Audit Metadata